Learning to build tech product in the era of AI, where anyone can build tech products.

This article is about the reflection of my experience on building tech product in the era of AI where anyone can build tech products using AI like claude code, opencode, and other coding agents.

I am writing this to remind myself that, the beginning phase of building products using coding tools like claude code and opencode can be fun and engaging, but it also brings some painful experiences.

Ever since i first got to know about opencode that was like around 6 months ago, it felt like wonderful. Previously i had used Cursor IDE, so i knew how to use AI coding tools, but the thing i felt wonderful about opencode was there are free models, like seriously free models with generous limit. I started using opencode, and it felt like using free models and doing work, what a wonderful tool.

But, using opencode frequently made me dependent on the AI coding tools. My workflow started to become like this. Chat with the claude or chatgpt about the idea, brainstorm with them, then get the PRD (Product Requirements Documents) at end of the chat session, then feed that docs to the opencode. Few of the initial small projects like Nepal's Vehicle Tax Calculator, visual representation of how the Apache Spark works were build good. Like one shot prompt, and there it is. This got me addicted to use coding tools more and more. My everyday workflow became talking to AI chatbots, and using AI to build things.

As time went on, more and more dependency on the opencode increased, and more of the expectations as well. Looking at the code and running them manually became non-existent. Even opening VS code became like old things. Every task, every idea, every coding job, every research was happening in the opencode. And, i don't mean to say the AI coding tools/ AI tools are bad. They are very good at productivity, working fast. But depending on the AI coding tool/ AI tool massively degrades the thinking capacity of the brain. The situation will arrive where thinking about solution for the problem/ issue/ bug felt like burden, the inner feeling was "just feed that problem/ issue/ bug to opencode".

Now, after few months of using opencode massively (like using opencode from morning to the evening, till the free tier token limit reached), I felt like my brain thinking capacity was terribly. After that, the job interviews gone bad because i was not able to think about the problem scenario asked in the interview. Almost 4 interviews had been bad, and the reason was I became too much dependent on AI coding agents/ tools. Also, the excess use of AI coding tools makes you feel like, I know this thing but if someone asks the small details, then you won't know it, because hands on coding is almost zero. In one of the interview, an interviewer asked me what embedding model did I use for the embedding generation? I knew this, cause prior to when i was not high dependent on opencode, i had used and called that embedding model many times, but with increase on using AI coding tools, I forgot the embedding model name that i used in project, and couldn't say during interview.

But, now here's a plan on how i would build tech products on the era of AI, where anybody could build one. I am having thoughts to completely ditch the AI coding tools and code by hand. This thought feels very courageous now, but it was normal before AI phase. Coding and building a product seems like same thing, isn't it? We code to build tech product, right? But, what I think is product needs user, and code is what we use to build a product. So, building a product is one thing, and marketing, finding users, collecting feed backs, reiterating on the feed backs, and making progress by each iteration is what it takes to build a product.

So, what i realized just some time ago was, instead of using AI coding tools to build a full system with proper backend, why not use the frontend (minimal but modern UI/UX design as per the product/ idea needed) and deploy them in like cloudflare or vercel, and build a wait list first and see the page visit analytics. I believe wait list and page visit metrics will show if people we use the product or are eager for the product or not.

That's what i am going to try from now. Instead of building backend, frontend, and deploying in the AWS, let's just first really know if people want the product or not, are they eager for the product or not. I believe this approach will enable lot of indie solo developers to save time, efforts, and work on idea/ product that will have real users. Otherwise, most of the ideas that developers are building would just be side project that will never reach to people.